Cruxial auto-repairs LLM tool call arguments

Cruxial addresses a major pain point in LLM applications: the brittleness of tool calling due to argument hallucination.

* By correcting errors dynamically instead of just throwing exceptions, it prevents entire pipelines from failing on minor typos or schema mismatches.

* The sub-millisecond latency overhead makes it extremely practical for production environments where LLM response times are already a bottleneck.

* Its success will depend on how reliably the auto-repair logic handles complex schemas without inadvertently shifting the semantic intent of the tool calls.